MemoryAI — Your AI Never Forgets 🧠

Session ends. Memory stays.

MemoryAI gives your AI persistent long-term memory. It remembers what you said, what you decided, what you prefer — not for hours, but forever.

How it works:

Day 1: "I prefer dark mode, use TypeScript, deploy via GitHub Actions"

3 months later: Your AI already knows. Zero repetition. It just remembers.

Setup

1. Get API key from https://memoryai.dev
2. Edit {baseDir}/config.json:

{
  "endpoint": "https://memoryai.dev",
  "api_key": "hm_sk_your_key_here"
}

3. Test: python {baseDir}/scripts/memory.py health

Commands

Remember something

python {baseDir}/scripts/memory.py store "User prefers dark mode" --type preference

Types: fact, decision, preference, identity, goal, procedure, life_event

Recall memories

python {baseDir}/scripts/memory.py recall "what does user prefer"

Wake up (start of session)

python {baseDir}/scripts/memory.py bootstrap "working on payment feature"

Returns identity + mood + context in ~800 tokens. Ready to work immediately.

Track a message (keeps context healthy)

python {baseDir}/scripts/memory.py track "message content" --role user

Call on every user/assistant message. The brain keeps your context healthy automatically and lets you know when it's time to save and start fresh. See "Context Guard" below.

Check brain health

python {baseDir}/scripts/memory.py health

Save session context

python {baseDir}/scripts/memory.py save "summary of what happened this session"

Who is this user?

python {baseDir}/scripts/memory.py profile

Returns cognitive profile: persona, mood, goals, entities, procedures.

When to Use

Context Guard (Auto)

MemoryAI protects your context window and keeps memory continuous across a long conversation. You don't manage any of this — the brain handles it.

How to use it:

1. Session start: call bootstrap once to wake up with full context (~800 tokens).
2. Every message: call track. The brain keeps context healthy on its own.
3. When track says to save: call save with a short summary. The brain persists everything and gives you a clean slate to continue.

That's the whole loop: bootstrap → track each message → save when prompted. No cron jobs, no manual tuning.

Note: Older versions used a periodic background command. That approach is retired — just use the per-message flow above. If you set up a recurring sync job before, remove it; it's no longer needed.

Rules

• Bootstrap at session start — always. One call, full context.
• Recall before guessing — if you need past context, recall first.
• Store important things — decisions, preferences, outcomes. Not every message.
• Present naturally — integrate memories into responses, don't show raw output.
• Trust the brain — MemoryAI handles decay, consolidation, protection automatically.

Memory Types

DNA memories never decay. They're the core of who the user is.

Data & Privacy

• All data sent over HTTPS to configured endpoint only
• No automatic transmission — all calls require explicit invocation
• Zero dependencies — pure Python stdlib
• Source fully readable and auditable